Journal of Memory and Language 46, 441—517 (2002)
doi:10.1006/jmla.2002.2864, available online at http://www.academicpre
The Nature of Recollection a
A Review of 30 Years of
Andrew P. Yonelina
D
T
pro o
eve a
the d
from s
dic d
me b
par o
fam ,
the n
enc d
neu w
sub
cor
exp
gle
Rec
ory
thre
pro
dic
Resu
and neu
creasin
perform
process
as reco
is illust
ognizin
to reco
were p
tions s
based
This re
Institute
Joanna S
Mark Ki
sky, Cha
ments on
Addre
P. Yoneli
fornia, DUniversity of California,
o account for dissociations observed in recognition memory
posed that assume that recognition judgments can be based
nts or on the assessment of stimulus familiarity. In the current
methods that have been developed to measure recollection an
behavioral, neuropsychological, and neuroimaging studies i
tions. Results from a variety of measurement methods, inclu
thods, are found to lead to remarkably consistent conclusions a
ticularly when ceiling effects are avoided. For example, rec
iliarity to response speeding, division of attention, generation
amnestic effects of benzodiazepines, but it is less sensitive tha
y manipulations, forgetting over short retention intervals, an
ropsychological and neuroimaging results indicate that the t441
strates and provide support for models that assume that recollec
tex, whereas familiarity relies on regions surrounding the hipp
erimental manipulations at time of test indicate that the two pr
dissociations produced by study manipulations indicate that th
ollection is similar but not identical to free recall, whereas fam
, but is dissociable from perceptual implicit memory. Finally, t
sholdlike retrieval process that supports novel learning, wh
cess that can support novel learning only under certain conditio
tions and prove useful in resolving several theoretical disagreem
lts from cognitive, neuropsychological,
roimaging studies of human memory in-
gly indicate that recognition memory
ance re ects two distinct memory
es or types of memory, often referred to
llection and familiarity. The distinction
rated by the common experience of rec-
g a person as familiar but not being able
llect who the person is or where they
reviously encountered. Such introspec-
uggest that memory judgments can be
either on recollection of information
about p
of stim
ment o
tive psy
distinct
has bee
Given
that are
framew
fully ex
the me
ure the
grate th
this art
ture an
tion an
models
Do w
ory or
ory per
perform
search was supported by a grant from the National
of Mental Health MH59325. I am grateful to
cheib, Neal Kroll, Michele Lazzara, Joel Quamme,
shiyama, Eric Nolan, Tonya Jacobs, Craik Brozin-
ran Ranganath, and the reviewers for their com-
earlier drafts of this article.
ss correspondence and reprint requests to Andrew
nas, Department of Psychology, University of Cali-
avis, CA 95616. E-mail: apyonelinas@ucdavis.edu.ss.com on
nd Familiarity:
Research
s
avis
tests, several dual-process models have been
n the recollection of details about previous
rticle, these models are examined, along with
familiarity. The relevant empirical literature
then reviewed in order to assess model pre-
ing task-dissociation and process-estimation
out the nature of recollection and familiarity,
llection is found to be more sensitive than
semantic encoding, the effects of aging, and
familiarity to shifts in response criterion, u-
some perceptual manipulations. Moreover,
o processes rely on partially distinct neural0749-596X/02 $35.00
' 2002 Elsevier Science (USA)
All rights reserved.
tion relies on the hippocampus and prefrontal
ocampus. Double dissociations produced by
ocesses are independent at retrieval, and sin-
ey are partially independent during encoding.
iliarity is similar to conceptual implicit mem-
he results indicate that recollection re ects a
ereas familiarity re ects a signal-detection
ns. The results verify a number of model pre-
ents. ' 2002 Elsevier Science (USA)
revious study events or on assessments
ulus familiarity. Although the develop-
f formal dual-process models by cogni-
chologists only began in the 1970s, the
ion between these two types of memory
n discussed since the time of Aristotle.
the rapid growth of empirical ndings
being interpreted within a dual-process
ork, there is an increasing need to care-
amine dual-process models, as well as
thods that have been developed to meas-
se processes, and to evaluate and inte-
e related empirical literature. The aim of
icle is to characterize the functional na-
d neuroanatomical substrates of recollec-
d familiarity by examining the existing
in light of the empirical evidence.
e need more than a single type of mem-
process to account for recognition mem-
formance? If it is possible to account for
ance with a model that involves only a

442 N
single t
should
memor
that at
are nee
tion m
conside
conclus
pirical
port for
than a
will als
mains
accumu
First
cated th
For exa
jects ar
crimina
such as
recently
quickly
require
about th
or whe
Hintzm
wards,
Levin,
nition t
tests, re
have sh
respons
ing a n
item, o
increas
curacy/
1984; G
Curren
Jacoby,
sults in
the rela
and on
jects ab
lows th
Seco
dence r
familia
ing cha
hit rate
o
s
e
he
en
(
&
,
in
t
o
e
a
iv
ct
li
&
im
a
R
ls
O
e
n
ts
r
ct
,
e
e
e
i
fr
o
c
in
,
u
r
o
raANDREW P. YONELI
ype of memory then this simpler model
be preferred. Although a majority of
y researchers today would likely agree
least two processes or types of memory
ded to account for the existing recogni-
emory literature, it is useful to brie y
r the type of results that have led to this
ion. I start by brie y describing four em-
dissociations that have been used as sup-
the claim that recognition involves more
single type of memory. These examples
o be useful in illustrating the general do-
from which the empirical evidence has
lated.
, studies of processing speed have indi-
at familiarity is faster than recollection.
mple, under speeded test conditions sub-
e found to be able to make accurate dis-
tions that can be based on familiarity,
distinguishing between items that were
studied and nonstudied items, more
than they can make discriminations that
them to recollect speci c information
e study event, such as determining when
re an item was previously studied (e.g.,
an & Caulton, 1997; Gronlund, Ed-
& Ohrt, 1997; Hintzman, Caulton, &
1998). I refer to these two types of recog-
ests as item and associative recognition
spectively. A number of related studies
own that as the time allowed to make a
e is increased, the probability of accept-
ew item that is either similar to a studied
r is from an inappropriate study list, rst
es then decreases, producing biphasic ac-
response-time functions (e.g., Dosher,
ronlund & Ratcliff, 1989; Hintzman &
, 1994; Jacoby, 1999; McElree, Dolan, &
1999; Rotello & Heit, 2000). These re-
dicate that a fast familiarity process leads
ted new items to be incorrectly accepted,
ly with additional retrieval time are sub-
le to recollect the information that al-
em to reject those items.
functi
the ob
and th
that t
indep
them
ford,
Reder
Dobb
indica
comp
tion p
tests,
the RO
sociat
distin
more
Kelly
& Sh
Van T
Qin,
mode
tion R
ciativ
are co
of tes
Thi
distin
ample
on th
that it
sociat
of the
hibit d
tions
nized
of re
Yonel
Smith
nas, K
Altho
brain
they d
rate bnd, the analysis of recognition con -
esponses indicates that recollection and
rity can produce distinct receiver operat-
racteristics (ROCs). For example, when
s are plotted against false alarm rates as a
Four
rupted
indicat
ent on
nesic pAS
n of response con dence (i.e., an ROC),
erved empirical functions are curvilinear
ir shapes change across conditions such
y require no less than two functionally
dent memory parameters to describe
e.g., Kelley & Wixted, 2001; Kim, Hil-
Adams, 1999; Rotello, Macmillan, &
2001; Yonelinas, 1994, 1999; Yonelinas,
s, Szymanski, Dhaliwal, & King, 1996),
ing that at least two separate memory
nents are needed to account for recogni-
rformance. Moreover, in item recognition
s long as performance is above chance,
Cs are always curvilinear, whereas in as-
e recognition tests the ROCs are quite
in the sense that they are often much
near in shape (e.g., Arndt & Reder, 2001;
Wixted, 2001; Slotnick, Klein, Dodson,
amura, 2000; Rotello, Macmillan, &
ssel, 2000; Yonelinas, 1997, 1999; but see
aye, Johnson, & Mitchell, 2001). Thus,
that provide a good t for item recogni-
Cs tend to provide a poor t of the asso-
ROCs, indicating that different processes
tributing to performance in the two types
.
d, recollection and familiarity exhibit
electrophysiological correlates. For ex-
event related potentials (ERPs) recorded
scalp during recognition tests indicate
ms that are remembered or that are as-
d with accurate memory for some detail
study event are related to ERPs that ex-
stinct temporal and spatial scalp distribu-
om those related to items that are recog-
n the basis of familiarity in the absence
ollection (e.g., Curran, 2000; D zel,
as, Mangun, Heinze, & Tulving, 1997;
1993; Klimesch, Doppelmayr, Yoneli-
roll, Lazzara, Rohm, & Gruber, 2001).
gh these results do not indicate which
egions support recognition performance,
indicate that there are at least two sepa-
in processes involved.th, recollection is more severely dis-
than familiarity by certain brain injuries,
ing that these two processes are depend-
different brain regions. For example, am-
atients exhibit signi cantly greater mem-